返回更新列表
已更新Aug 31, 2026

Awesome-LLMs-for-Vulnerability-Detection — 已更新!

社区最全面、持续更新的关于大语言模型用于软件漏洞检测的研究索引——涵盖函数级、仓库级、智能体及智能合约检测的论文，以及数据集、基准测试和综述。

分享

用于漏洞检测的大型语言模型精选列表

一份关于使用LLM进行漏洞检测与发现的论文、项目和智能体技能的精选列表。

📄 论文

仅展示2025年及以后的工作。更早的工作请参阅论文存档（2024年及以前）。

标题会议/期刊年份论文Github
VulnGym: Benchmarking Coding Agents for Repository-Level Vulnerability Detection2026链接链接
VulTriage: Triple-Path Context Augmentation for LLM-Based Vulnerability Detection2026链接链接
Synthesizing Multi-Agent Harnesses for Vulnerability Discovery2026链接链接
QRS: A Rule-Synthesizing Neuro-Symbolic Triad for Autonomous Vulnerability Discovery2026链接
Seclens: Role-specific Evaluation of LLM's for security vulnerablity detection2026链接链接
Do Fine-Tuned LLMs Understand Vulnerabilities? An Investigation into the Semantic Trap2026链接
Sifting the Noise: A Comparative Study of LLM Agents in Vulnerability False Positive FilteringISSTA2026链接
AgenticSCR: An Autonomous Agentic Secure Code Review for Immature Vulnerabilities Detection2026链接
LLM-based Vulnerability Detection at Project Scale: An Empirical Study2026链接
MulVul: Retrieval-augmented Multi-Agent Code Vulnerability Detection via Cross-Model Prompt Evolution2026链接
VulnLLM-R: Specialized Reasoning LLM with Agent Scaffold for Vulnerability Detection2025链接
VULPO: Context-Aware Vulnerability Detection via On-Policy LLM Optimization2025链接
VulInstruct: Teaching LLMs Root-Cause Reasoning for Vulnerability Detection via Security Specifications2025链接
From Large to Mammoth: A Comparative Evaluation of Large Language Models in Vulnerability DetectionNDSS2025链接
Benchmarking LLMs and LLM-based Agents in Practical Vulnerability Detection for Code RepositoriesACL2025链接链接
A Systematic Literature Review on Detecting Software Vulnerabilities with Large Language Models2025链接链接
LLMxCPG: Context-Aware Vulnerability Detection Through Code Property Graph-Guided Large Language ModelsUsenix2025链接链接
CLeVeR: Multi-modal Contrastive Learning for Vulnerability Code RepresentationACL Findings2025链接链接
Mono: Is Your "Clean" Vulnerability Dataset Really Solvable? Exposing and Trapping Undecidable Patches and Beyond2025链接链接
Learning to Focus: Context Extraction for Efficient Code Vulnerability Detection with Language Models2025链接
SV-TrustEval-C: Evaluating Structure and Semantic Reasoning in Large Language Models for Source Code Vulnerability AnalysisSP2025链接链接
SecVulEval: Benchmarking LLMs for Real-World C/C++ Vulnerability Detection2025链接链接
CVE-Bench: Benchmarking LLM-based Software Engineering Agent's Ability to Repair Real-World CVE VulnerabilitiesNAACL2025链接链接
R2Vul: Learning to Reason about Software Vulnerabilities with Reinforcement Learning and Structured Reasoning Distillation2025链接链接
Neuro-symbolic Static Analysis with LLM-generated Vulnerability Patterns2025链接
Context-Enhanced Vulnerability Detection Based on Large Language Model2025链接
Everything You Wanted to Know About LLM-based Vulnerability Detection But Were Afraid to Ask2025链接链接
MOS: Towards Effective Smart Contract Vulnerability Detection through Mixture-of-Experts Tuning of Large Language Models2025链接
Abundant Modalities Offer More Nutrients: Multi-Modal-Based Function-Level Vulnerability DetectionTOSEM2025链接链接
Generative Large Language Model usage in Smart Contract Vulnerability Detection2025链接
Closing the Gap: A User Study on the Real-world Usefulness of AI-powered Vulnerability Detection & Repair in the IDEICSE2025链接链接
Vulnerability Detection with Code Language Models: How Far Are We?ICSE2025链接链接
Combining Fine-Tuning and LLM-based Agents for Intuitive Smart Contract Auditing with JustificationsICSE2025链接
LAMD: Context-driven Android Malware Detection and Classification with LLMs2025链接
LLMs in Software Security: A Survey of Vulnerability Detection Techniques and Insights2025链接链接
One-for-All Does Not Work! Enhancing Vulnerability Detection by Mixture-of-Experts (MoE)2025链接

🚀 项目

名称描述Github
OpenAnt (Knostic)基于LLM的多阶段漏洞发现，具备对抗性验证能力链接
DeepAudit多智能体AI红队平台，支持Docker沙箱漏洞利用验证链接
AutoCVE采用多智能体架构的自动化漏洞检测与报告工具链接
Darkmoon开源（GPL-3.0）自主AI渗透测试平台及MCP主机；按技术分类的进攻性子智能体，覆盖Active Directory和Kubernetes，80+编排工具，每个发现均留有证据链链接
strix通过pip安装的开源自主AI渗透测试工具链接
deepsec (Vercel)面向深度代码库漏洞扫描的安全框架，支持编码智能体链接

🧩 智能体技能

名称描述链接
codex-security (OpenAI)通过Codex智能体实现自主仓库级漏洞扫描链接
defending-code-reference-harness (Anthropic)使用Claude Code进行威胁建模、扫描、分类和修补的参考技能链接
security-audit-skill (Cloudflare)六阶段安全审计技能，配备并行狩猎智能体和对抗性验证链接

📡 arxiv.md

通过工作流自动每日捕获并更新指定关键词的Arxiv论文。

致谢

本项目的Updated Arxiv Papers Daily工作流借鉴了LLM4SE项目。我通过使用arxiv库对其原始代码进行了重构。

阅读更多

分类

静态分析漏洞分析代码分析机器学习论文与研究精选资源AI 安全